A graph based preordonnances theoretic supervised feature selection in high dimensional data
作者：
Highlights：
• Two novel measures are defined to evaluate the relevance and redundancy of each predictor of any type.
• A new hybrid filter–wrapper feature selection approach is proposed to select the most important features.
• The filter phase is based on a novel feature evaluation criterion (MaCΨ weight) that is related to the defined relevance and redundancy measures simultaneously.
• The wrapper phase is based on graph theory and also on sequential backward selection.
摘要
•Two novel measures are defined to evaluate the relevance and redundancy of each predictor of any type.•A new hybrid filter–wrapper feature selection approach is proposed to select the most important features.•The filter phase is based on a novel feature evaluation criterion (MaCΨ weight) that is related to the defined relevance and redundancy measures simultaneously.•The wrapper phase is based on graph theory and also on sequential backward selection.
论文关键词：Preordonance,Relevance,Redundancy,Maximal clique,Selection,High-dimensional dataset
论文评审过程：Received 10 February 2022, Revised 5 August 2022, Accepted 13 September 2022, Available online 21 September 2022, Version of Record 1 October 2022.
论文官网地址：https://doi.org/10.1016/j.knosys.2022.109899
